| name | analyze-customer-feedback |
| category | data |
| description | Synthesize interviews, surveys, reviews, support records, and sales notes into traceable themes, counterevidence, segments, and product decisions. Use when feedback is scattered, stakeholders quote anecdotes selectively, or a team needs to understand recurring needs without flattening context. |
analyze-customer-feedback
Preserve the voice and context of each record while looking for patterns. Produce a synthesis report
where every theme is traceable, counted appropriately, and paired with evidence that challenges it.
Inputs
- Name the decision, sources, time window, eligible population, owner, and privacy restrictions.
- Gather raw records with source, date, segment, channel, and consent or usage basis.
- Record collection bias and groups missing from the material.
Procedure
- Define the research questions before reading for themes.
- Normalize metadata without rewriting participant language. Remove duplicates and mark linked records.
- Redact unnecessary personal data and separate restricted raw material from shareable excerpts.
- Read a varied sample and create a codebook with definition, inclusion, exclusion, and examples.
- Code each record with one or more themes, sentiment only when useful, journey stage, severity, and requested outcome.
- Review ambiguous records and revise the codebook. Recode earlier material when definitions change.
- Count records, people, accounts, and sources separately so a noisy channel or frequent reporter cannot dominate.
- Compare themes across relevant segments and time. Preserve counterexamples and conflicting needs.
- Connect themes to behavioral or operational evidence when available; do not treat stated preference as observed use.
- Write implications as decisions or hypotheses with confidence, affected users, evidence, and next test.
- Publish a synthesis with traceable anonymous record ids and documented limitations.
Boundaries
Do not use private support or sales content outside its allowed purpose. Never infer protected traits,
present sentiment software as ground truth, or turn a few vivid quotes into population claims.
Done
- A codebook defines each theme with inclusion, exclusion, and examples
- A coded dataset preserves source, segment, date, traceable record id, and privacy handling
- The synthesis reports counts by the right unit, counterevidence, collection bias, and missing groups
- Each recommendation names supporting evidence, confidence, decision owner, and next verification step
Then use conduct-user-interviews for unresolved context or define-product-metrics for measurable follow-up.
